Outcast 1.1 (1999/2014) | 4K | RTX 3090 | 5950X
1Without recording add up ~2-5% of performance when gpu is at 95-100% usage.
Timestamps
00:00 Settings
00:36 Max voxel CPU thread comparison (8 vs default - 32)
01:13 Gameplay
For recording i used Nvidia Shadowplay.
Shadowplay settings:
3840x2160 resolution;
Framerate: 60;
Bitrate: 70mbps.
PC info:
cpu: AMD Ryzen™ 9 5950X;
gpu:GeForce RTX™ 3090 GAMING OC;
ram: G Skill TridentZ 64gb DDR4 3200 Mt/s (4 sticks, dual channel, quad rank)
tCL16, tRCDWR15, tRCDRD15, tRP15, tRAS34, tRC75, tRFC (ns)250, tRFC400;
mobo: ROG Crosshair VIII Hero X570 (WI-FI);
soundcard: Creative Sound BlasterX Ae-5 ;
cpu cooler: NH-D15 chromax.black;
pc case: Corsair 780T;
psu: Corsair HX1200 Platinum 1200W;
Operating system: Windows 10 Pro.
Game Drive - HDD (WDC WD4003FRYZ) 7200RPM, 256mb cache, CMR, 4TB.
This is the drive you see on my RTSS performance overlay and it is being used solely for games, for video recording i use different hard drive.
The "HDD" usage statistic you see on my overlay is ONLY from the game running and is not effected by any other application.
Boot Drive - sata SSD, Samsung 850 EVO 512GB.
All 5950x settings were set on AUTO.
All 5950X settings were set on AUTO (in other words completely stock), amd virtualization feature has been enabled in the bios, nvidia resizeable BAR is enabled.
My motherboard's AUTO disables PBO by default. Power limits set to its defaults
PPT - 142W;
TDC - 95A;
EDC - 140.
Some explanation of my overlay statistics:
"GPU PERF LIMIT" also known as "PerfCap Reason". As name suggests, shows the reason, why GPU performance is capped, limited. If reason becomes "no load" it means that gpu itself is not the limiting factor.
"(INSTANT) CLOCKSPEED" Is the most popular cpu clockspeed measurement method, found in most PC component statistic software, EG. MSI Afterburner, FPS monitor, RTSS, Etc.
This method ignores cpu core sleeping state and it is determined as RATIO * BUS Clock at the polling edge during active state. In other words it is highest measured value during cpu active state and it is highly innacurate on modern cpu's, due to very quick auto boosting capabilities, algorithm and sleeping states.
GREY NUMBERS INSIDE EACH CPU THREAD BARCHART - These values are "EFFECTIVE CLOCKSPEED" most commonly found in application such as "HWiNFO". This is the most accurate method of measuring cpu core speed. It is simply an average clockspeed calculated across the polling interval with respect to sleeping states. To have a better understanding why it is important you need to understand that modern cpus can enter sleep state at extremely fast pace, every time, your cpu core does not have task assigned to it, it enters sleep mode to save on power consumption, current and temperatures.
"GPU Busy" is a new feature presented by intel, which measures GPU active time in rendering pipeline. In perfect scenario GPU busy time should match game frametimes (FT in my OSD), that means the GPU is fully ocuppied in game rendering, in other words - fully utilized and optimized without any external stalls, such as driver overhead, cpu bottleneck, ram bottleneck.
PS. For GPU busy statistic i used 3 second delay for smoother graph data delivery.
"Bottleneck: CPU/GPU" analyzes data stream from statistics of "msGpuActive" or "GPU Busy" and "msBetweenPresents" or "Game frametimes(FT)" and reports if "GPU busy" time takes 80% or more of frametime.
"GAME VRAM USAGE" also known as "Process VRAM usage" - shows how much vram current running process (in this case, the game) requires, others likes to call it "dedicated vram usage" even though it is bad term to use.
"TOTAL ALLOCATED VRAM" - shows how much of VRAM is allocated from all programs in windows, windows itself, if opened - web browsers, spotify, discord, etc. Also allocation increases from multiple monitors as well, so the more monitors and with higher resolution are connected the higher allocation will be.
(This value is the same as in windows task manager, under GPU tab "Dedicated GPU memory usage")
Regarding the whole VRAM usage debate, none of these two values shows true game VRAM requirements and when it will start to starve of VRAM deficiency. And there are no way to truly know it unless you test with less VRAM, but usually its impossible to test with same GPU as there are no ways to artificially limit VRAM usage. These values are more suited as guidelines rather than definitive proof, but usually performance starts to drop, when game has saturated 100% of "Total allocated VRAM" and it will have issues when "Process VRAM usage" will be at 100% of total GPU VRAM buffer.
The monitoring program used in the top left corner is MSI Afterburner On Screen Display + Rivatuner Statistic Server. The actual overlay was displayed using "RTSS overlay editor" and not from MSI afterburner.